2toINF/X-VLA-WidowX
Partial fine-tune vs microsoft/Florence-2-large. Partial change: 90% of the weights changed (attention, embeddings, mlp, norms, other), in 12 of 12 layers.
Tensors changed: Embeddings 2/6, Norms 154/154, other 104/106, MLP 144/144, Attention 192/192